TECHNOLOGY, COMPUTING, AND SIMULATION SOCIETY FOR TECHNOLOGY IN ANESTHESIA

SECTION EDITOR
STEVEN J. BARKER

Predictive Performance of Three Multivariate Difficult

Tracheal Intubation Models: A Double-Blind,
Case-Controlled Study
Mohamed Naguib, MB, BCh, MSc, FFARCSI, MD*, Franklin L. Scamman, MD‡,
Cormac O’Sullivan, CRNA‡, John Aker, CRNA§, Alan F. Ross, MD‡,
Steven Kosmach, MSN, RN*, and Joe E. Ensor, PhD†
Departments of *Anesthesiology and Pain Medicine and †Biostatistics and Applied Mathematics, The University of Texas
M. D. Anderson Cancer Center, Houston; ‡Department of Anesthesia, The University of Iowa Roy J. and Lucille A.
Carver College of Medicine, Iowa City; and §Department of Anesthesia, Children’s Mercy Hospitals & Clinics, Kansas
City, Missouri

We performed a case-controlled, double-blind study to significantly more sensitive (81.4%; P ⬍ 0.0001) than the
examine the performance of three multivariate clinical Arné (54.6%) or Wilson (40.2%) models. Both the Naguib
models (Wilson, Arné, and Naguib models) in the predic- (76.8%) and Arné (74.7%) model classified more intuba-
tion of unanticipated difficult intubation. The study tions correctly (P ⫽ 0.01) than the Wilson model (66.5%).
group consisted of 97 patients in whom an unanticipated The specificity of Arné, Wilson, and Naguib model was
difficult intubation had occurred. For each difficult intu- 94.9%, 92.8%, and 72.2%, respectively (P ⬍ 0.0001). The
bation patient, a matched control patient was selected in corresponding area under the receiver operating charac-
whom tracheal intubation had been easily accomplished. teristic curve was 0.87, 0.79, and 0.82, respectively. Our
Postoperatively, a blinded investigator evaluated both new model for prediction of difficult intubation was de-
patients. The clinical assessment included the patient’s veloped using logistic regression and includes thyromen-
weight, height, age, Mallampati score, interincisor gap, tal distance, Mallampati score, interincisor gap, and
thyromental distance, thyrosternal distance, neck circum- height. This model is 82.5% sensitive and 85.6% specific
ference, Wilson risk sum score, history of previous diffi- with an area under the receiver operating characteristic
cult intubation, and diseases associated with difficult la- curve of 0.90.
ryngoscopy or intubation. The Naguib model was (Anesth Analg 2006;102:818 –24)

U
nanticipated difficult tracheal intubation is a sig- Britain for 2000 –2002, 3 of the 6 deaths directly attrib-
nificant source of morbidity and mortality in utable to anesthesia were associated with failed tra-
anesthetic practice. The incidence of difficult in- cheal intubation (8). Worldwide, up to 600 patients are
tubation in the operating room has been reported to thought to die annually as a result of complications
range from 1% to 18% (1– 4). The incidence of occurring at the time of tracheal intubation (9). Ap-
abandoned/failed intubation is approximately 0.05%– proximately 30% of the deaths in patients who expe-
0.35% (5,6), whereas that of cannot ventilate by mask, rienced difficulties at laryngoscopy or intubation are
cannot intubate is around 0.0001%– 0.02% (7). In the caused by hypoxic brain damage secondary to inabil-
Confidential Enquiry into Maternal Deaths in Great ity to maintain a patent airway (2).
Difficult tracheal intubation accounted for approxi-
mately 17% of adverse respiratory events in an Ameri-
can Society of Anesthesiologists closed-claims analysis
Presented, in part, at the American Society of Anesthesiologists (10). In 85% of these cases, the outcome was either death
Annual Meeting, October 22–26, 2005, New Orleans, LA.
Accepted for publication October 17, 2005. or brain damage (10). Increases in the incidence of mor-
Address Correspondence and reprint requests to Mohamed bid nonfatal events have also been noted in patients who
Naguib, MB, BCh, MSc, FFARCSI, MD, Department of Anesthesi- have undergone difficult tracheal intubation (11–14).
ology and Pain Medicine, Unit 409, The University of Texas M. D.
Anderson Cancer Center, 1400 Holcombe Blvd., Houston, TX 77030. These events included desaturation, hypertension,
Address e-mail to Naguib@mdanderson.org. esophageal intubation, pharyngeal trauma, dental in-
DOI: 10.1213/01.ane.0000196507.19771.b2 jury, cancellation of surgery, increased hospital stay, and

©2006 by the International Anesthesia Research Society

818 Anesth Analg 2006;102:818–24 0003-2999/06
ANESTH ANALG TECHNOLOGY, COMPUTING, AND SIMULATION NAGUIB ET AL. 819
2006;102:818–24 COMPARISON OF DIFFICULT TRACHEAL INTUBATION MODELS

an increased rate of unexpected intensive care unit ad- Lehane (5), Grade 3 means that, during laryngoscopy,
mission. none of the glottis, but part or the entire epiglottis, can
In most studies, difficult laryngoscopy has been be seen, whereas Grade 4 means that neither the la-
defined as a view of the larynx corresponding to grade ryngeal structures nor the epiglottis can be visualized.
3 or 4 in the classification of difficult intubation by Postoperatively, patients who had unanticipated
Cormack and Lehane (5). The American Society of difficult intubation were approached by an investiga-
Anesthesiologists defines difficult tracheal intubation tor after they had fully awoken from general anesthe-
as when “proper insertion of the endotracheal tube sia. If a patient agreed to participate in the study, an
with conventional laryngoscopy requires more than 3 investigator invited a second patient from that day’s
attempts, or more than 10 min” (15). surgical schedule to participate as a control. Each con-
Although unanticipated difficult intubation has trol patient who had undergone uneventful general
been the subject of many studies, a puzzling feature of anesthesia without any reported difficulties at laryn-
these studies is the wide variation in the reported goscopy or tracheal intubation was closely matched
sensitivity of the different models used for prediction demographically with a study patient to age, weight,
of this problem (1,3,4,6,16 –22). A test performed to height, and sex. Difficult intubation patients and their
predict difficult intubation should have high sensitiv- matched controls were the only patients who con-
ity so that it will identify most patients in whom sented to participate in this study. A second blinded
intubation will truly be difficult. We are not aware of investigator then evaluated the two patients in the
any studies that have evaluated different multivariate postanesthesia care unit, second-stage recovery facil-
models in the same population of patients to deter- ity, or ward. To reduce measurement bias, patients
mine the most sensitive model for predicting difficult were instructed by the consenting investigator not to
intubation. Therefore, we designed and performed the comment on their sore throat, potential airway diffi-
double-blind, case-controlled study described herein culty, or any other aspect of their anesthetic experi-
to compare and validate the predictive performance of ence to the blinded investigator. The details of the
three multivariate clinical models described by Wilson laryngoscopic findings and degree of difficulty of in-
et al. (1), Arné et al. (20), and Naguib et al. (21) in a tubation were not known by the investigator who
group of patients who had confirmed unanticipated interviewed and measured the patient pairs. All of
difficult intubation. The sensitivity reported for the these assessments were performed by one of three
latter two models has been the highest sensitivity for investigators.
such clinical models reported in the literature. Subse- The clinical assessment included:
quently, we developed a new model for predicting
difficult intubation. 1. Measurement of weight, height, and recording of
age
2. Assessment of the airway according to the pha-
ryngeal structures seen by using the method de-
Methods scribed by Mallampati et al. (16) with the modi-
This protocol was approved by the IRB of the Univer- fication described by Samsoon and Young (6):
sity of Iowa Hospitals and Clinics, Iowa City, IA, and Class 1: soft palate, fauces, uvula, and pillars
each patient gave his or her written informed consent visible
to participate in the study. Adult patients presenting Class 2: soft palate, fauces, and uvula visible
to the University of Iowa Hospitals and Clinics for Class 3: soft palate and base of uvula visible
general anesthesia for any type of nonemergency sur- Class 4: none of the soft palate visible
gical procedures except traumatic facial abnormalities,
obstetric surgery, or cardiac surgery from October 3. Measurement of interincisor gap (in centimeters)
1999 to November 2004 were enrolled. Unanticipated with the mouth fully open
difficult intubation was identified by an experienced 4. Measurement of thyromental distance (in centi-
laryngoscopist (⬎5 yr in anesthetic practice after com- meters) along a straight line from the thyroid
pletion of training). For the purposes of this study, notch to the lower border of the mandibular
unanticipated difficult intubation was defined as dif- mentum with the head fully extended and mouth
ficult laryngoscopy (corresponding to a Grade 3 or 4 closed
Cormack and Lehane laryngoscopic view) and diffi- 5. Measurement of thyrosternal distance along a
cult tracheal intubation (2 or more attempts at placing straight line from the thyroid notch to the upper
the endotracheal tube) or the use of an alternative border of the manubrium sterni with the head
device (laryngeal mask airway [LMA; Laryngeal Mask fully extended and mouth closed
Company, Henley-on-Thames, United Kingdom] or 6. Measurement of neck circumference
bougie) when using optimal head and neck position- 7. Determination of the Wilson risk sum score (1),
ing (the sniff position). As defined by Cormack and which scores 5 factors (weight, head and neck
820 TECHNOLOGY, COMPUTING, AND SIMULATION NAGUIB ET AL. ANESTH ANALG
COMPARISON OF DIFFICULT TRACHEAL INTUBATION MODELS 2006;102:818–24

Table 1. Wilson Risk Sum Score (1) Table 2. Simplified Score Model Described by Arné et al.
(20) for Prediction of Difficult Intubation
Risk factor Level Variable
Risk factor Score
Weight 0 ⬍90 kg
1 90–110 kg Previous knowledge of difficult intubation
2 ⬎110 kg No 0
Head and neck 0 ⬎90º Yes 10
movement 1 About 90º (i.e., ⫾10º) Diseases associated with difficult intubation
2 ⬍90º No 0
Jaw movement 0 IG ⱖ5 cm or SLux ⬎ 0 Yes 5
1 IG ⬍ 5 cm and SLux ⫽ 0 Clinical symptoms of airway pathology
2 IG ⬍ 5 cm and SLux ⬍ 0 No 0
Receding mandible 0 Normal Yes 3
1 Moderate IG and mandible subluxation
2 Severe IG ⱖ 5 cm or SLux ⬎ 0 0
Buck teeth 0 Normal IG ⬍ 5.0–3.5 cm and SLux ⫽ 0 3
1 Moderate IG ⬍ 3.5 cm and SLux ⬍ 0 13
2 Severe Thyromental distance
ⱖ 6.5 cm 0
IG ⫽ Interincisor gap; SLux ⫽ Subluxation (maximal forward protrusion
of the lower incisors beyond the upper incisors). ⬍ 6.5 cm 4
Maximum range of head and neck movement
More than 100º 0
movement, jaw movement, receding mandible, About 90º (⫾10º) 2
Less than 80º 5
and buck teeth) from 0 to 2 for a total range of
Mallampati score
0 –10 (Table 1) Class 1 0
8. Recording history of previous difficulty with la- Class 2 2
ryngoscopy or intubation Class 3 6
9. Recording diseases associated with difficulties in Class 4 8
laryngoscopy or intubation, such as acromegaly Total possible 48
(23) and cervical spondylosis with limitation of IG ⫽ interincisor gap; SLux ⫽ subluxation (maximal forward protrusion of
neck movements (24). the lower incisors beyond the upper incisors).

The sensitivity and specificity of the models de-

scribed by Wilson et al. (1) (Table 2), Arné et al. (20) score was dichotomized such that a score of 1 or 2 was
(Table 2), and Naguib et al. (21) were assessed based scored as 0 and a score of 3 or 4 was scored as 1.
on the data collected. The criteria used to predict The receiver operating characteristic (ROC) curve
difficult tracheal intubation were Wilson risk sum was used to describe the discrimination abilities and
score ⱖ4 or more (1), Arné model score ⬎11 (20), and to explore the trade-offs between the sensitivity and
Naguib model score ⬍0 (21). The Naguib model is specificity of the different models (25). The ROC area
based on the formula clinical prediction ⫽ 4.9504 ⫹ under the curve (AUC) is frequently viewed as a ro-
(thyrosternal distance ⫻ 1.1003) ⫹ (Mallampati score bust indicator of the performance of classification
⫻ ⫺2.6076) ⫹ (thyromental distance ⫻ 0.9684) ⫹ (neck models. The AUC is a performance indicator equiva-
circumference ⫻ ⫺0.3966). lent to the nonparametric concordance measure, Som-
Positive predictive value and negative predictive ers D, and the difference between two ROC areas is
value were calculated based on a prevalence of diffi- half the difference between the corresponding Somers
cult intubation of 5.8% (4), as reported in recent meta- D values (26). The STATA software program (version
analysis. 8; Stata Corp, College Station, TX) was used to assess
Statistical analyses were performed by using the the difference between ROC AUCs based on the ␹2 test
SAS software program (version 9.1; SAS Institute Inc., developed from the generalized U-statistics theory by
Cary, NC). Each model was assessed based on the DeLong et al. (27).
entire group of patients. Thus, the Cochran Q value
was computed to test the homogeneity of the patient
groups. Demographic differences were determined by Results
using the ␹2 test and were considered significant when Data were collected on a convenience sample of all
P ⬍ 0.05. patients who presented to the operating rooms at the
We also subjected patient data (age, weight, height, University of Iowa Hospitals and Clinics between Sep-
sex, thyromental distance, Mallampati score, interin- tember 1999 and November 2004. During the study
cisor gap, and neck circumference) to a logistic regres- period, 210 patients were identified and consented to
sion model to identify variables that are predictors of participate in this study; 194 were included in the final
difficult intubation. For this analysis, the Mallampati analysis (97 with a difficult airway and 97 controls).
ANESTH ANALG TECHNOLOGY, COMPUTING, AND SIMULATION NAGUIB ET AL. 821
2006;102:818–24 COMPARISON OF DIFFICULT TRACHEAL INTUBATION MODELS

Table 3. Univariate Analysis of Clinical Variables

Difficult laryngoscopy and
intubation group Control group
(n ⫽ 97) (n ⫽ 97) P-value
Age, yr (mean ⫾ sd) 54.8 ⫾ 14.2 52.6 ⫾ 16.4 0.32
Sex ratio (male/female) 58/39 52/45 0.38
Weight, kg (mean ⫾ sd) 88.6 ⫾ 19.0 85.9 ⫾ 22.2 0.38
Height, cm (mean ⫾ sd) 170.9 ⫾ 12.2 170.4 ⫾ 12.2 0.79
IG, cm (mean ⫾ sd) 3.6 ⫾ 0.7 4.6 ⫾ 0.9 ⬍0.0001
Thyromental distance, cm (mean ⫾ sd) 6.86 ⫾ 1.3 7.95 ⫾ 1.00 ⬍0.0001
Thyrosternal distance, cm (mean ⫾ sd) 8.0 ⫾ 1.5 8.7 ⫾ 2.0 0.01
Neck circumference, cm (mean ⫾ sd) 42.5 ⫾ 4.9 40.9 ⫾ 4.7 0.03
Number of patients per Mallampati score (%)
Class 1 7 (7.2) 51 (52.6)
Class 2 25 (25.8) 39 (40.2) ⬍0.0001
Class 3 54 (55.7) 4 (4.1)
Class 4 11 (11.3) 3 (3.1)
IG ⫽ interincisor gap.

Sixteen patients (eight with a difficult airway and The highest sensitivity was achieved with the
eight controls) were excluded because of early dis- Naguib model (Table 4). Specifically, the sensitivity of
charge, incomplete data, or exclusion criteria. The total this model was 81.4% (95% confidence interval [CI],
patient sample yielded 80% power to detect significant 74.0%– 89.0%) compared with 40.2% (95% CI, 30.0%–
differences in model accuracy (odds ratio, 2.209; 30% 50.0%) for the Wilson model and 54.6% (95% CI,
model discordance) with the use of a two-sided Mc- 45.0%– 65.0%) for the Arné model. Naguib model was
Nemar test with a significance level of 0.05. significantly more sensitive than the other 2 models
There were no significant differences in the mean based on a pair-wise comparison using the McNemar
age, weight, or height between the two groups (Table test (P ⬍ 0.0001). Cochran Q statistic value indicated
3). However, the mean interincisor gap, thyromental that the 3 models differed significantly with respect to
distance, thyrosternal distance, neck circumference, their prediction accuracy (P ⬍ 0.02). Both Naguib
and Mallampati score differed significantly. model and Arné models classified more intubations
In the 97 patients in the difficult intubation group, correctly (P ⫽ 0.01) than the Wilson model (Table 5).
tracheal intubation was achieved under direct laryn- The McNemar test indicated that the Arné model and
goscopy after several attempts (mean ⫾ sd, 3.3 ⫾ 1.1) Naguib model did not differ significantly in their pre-
in 40 patients and with the use of a gum-elastic bougie diction accuracy (P ⫽ 0.6; kappa ⫽ 0.2). However, the
in another 19 patients. Direct laryngoscopic intubation specificity of the Arné model (94.9% [95% CI, 90.0%–
was completely unsuccessful in 38 of 97 patients. In 15 99.0%]) and Wilson model (92.8% [95% CI, 88.0%–
of these 38 patients in whom direct laryngoscopic 98.0%]) was significantly higher (P ⬍ 0.0001) than that
intubation was unsuccessful, fiberoptic-guided tra- of the Naguib model (72.2% [95% CI, 63.0%– 81.0%).
cheal intubation was performed successfully (9 while The ROC AUC that measured the discriminating
the patient was awake [awakened after intubation power of the Arné, Naguib, and Wilson model was
failed] and 6 while the patient was asleep, including 1 0.87 (95% CI, 0.82– 0.92), 0.82 (95% CI, 0.76 – 0.88), and
who underwent fiberoptic-guided tracheal tube place- 0.79 (95% CI, 0.72– 0.85), respectively (Fig. 1). The ROC
ment via an intubating LMA). Fiberoptic-guided intu- AUC for the Arné model was significantly greater
bation was unsuccessful in another five patients. In the than that of the Wilson model (P ⫽ 0.001).
remaining 23 of 38 patients, tracheal intubation was Logistic regression analysis identified four risk fac-
performed with the aid of an intubating LMA in 9 tors correlated with the prediction of difficult laryn-
patients. A LMA was used in another nine patients goscopy and intubation: thyromental distance, inter-
and blind nasal intubation in four patients. One pa- incisor gap, height, and Mallampati score. The
tient in whom tracheal intubation was difficult was prediction (l) was determined by the equation
allowed to awaken, and a regional technique was
used. This patient suffered postoperative oral trauma l ⫽ 0.2262 ⫺ 0.4621 ⫻ thyromental distance
and swelling. No other complications were noted. ⫹ 2.5516 ⫻ Mallampati score ⫺ 1.1461
The number of patients enrolled in the study during ⫻ interincisor gap ⫹ 0.0433 ⫻ height,
the first 15 months from September 1999 through De-
cember 2000 was 86 and decreased thereafter to 32, 18, in which the thyromental distance, interincisor gap,
30, and 28 patients in December 2001, December 2002, and height were measured in centimeters and Mal-
December 2003, and November 2004, respectively. lampati score was 0 or 1. Using this equation for
822 TECHNOLOGY, COMPUTING, AND SIMULATION NAGUIB ET AL. ANESTH ANALG
COMPARISON OF DIFFICULT TRACHEAL INTUBATION MODELS 2006;102:818–24

Table 4. Sensitivity, Specificity, Positive Predictive Value, and Negative Predictive Value of the Three Models Tested
Positive Negative
Model Sensitivity (%) Specificity (%) predictive value (%) predictive value (%)
Wilson model 40.2 (95% CI, 30.0–50.0) 92.8 (95% CI, 88.0–98.0) 25.6 96.2
Arné model 54.6 (95% CI, 45.0–65.0) 94.9 (95% CI, 90.0–99.0) 39.7 97.1
Naguib model 81.4 (95% CI, 74.0–89.0)* 72.2 (95% CI, 63.0–81.0)* 15.3 98.4
The positive and negative predictive values were calculated based on an overall incidence of difficult intubation of 5.8%, as reported in a recent meta-analysis
(4). CI ⫽ confidence interval. *P ⬍ 0.0001 as compared with the Wilson and Arné models.

Table 5. Prediction Accuracy of the Three Models Tested (95% CI, 73%– 89%), 85.6% (95% CI, 77%–91%), 26.1%,
Model Prediction Number Percentage and 98.8%, respectively. The ROC AUC for this model
was 0.90 (95% CI, 0.86 – 0.95) (Fig. 1).
Wilson model Correct 129 66.5
Wrong 65 33.5
A variable correlation analysis showed that height
Arné model Correct 145 74.7 was significantly correlated with both interincisor gap
Wrong 49 25.3 (P ⬍ 0.0001) and thyromental distance (P ⫽ 0.0007).
Naguib model Correct 149 76.8 The existence of this multi-colinearity allows height,
Wrong 45 23.2 which is not significant at the univariate level (Table
Cochran Q statistic value indicated that the Naguib and Arné models were 5), to be a significant factor in the multivariate model
significantly more accurate (P ⫽ 0.01) than Wilson model. (Table 6).
The total number of adult patients who underwent
general anesthesia and were initially eligible for the
study during the study period was 73,696. The trachea
proved unexpectedly difficult to intubate in 97 pa-
tients (0.13%) and was impossible to intubate in 38
patients (0.05%).

Discussion
This is the first validation study to evaluate different
multivariate models in the same population of pa-
tients in determining the most sensitive model for
prediction of difficult intubation. There was a substan-
tial difference in sensitivity and specificity among the
three models tested. The results show that the multi-
variate model described by Naguib et al. (16) is the
most sensitive (P ⬍ 0.0001) in identifying patients with
unanticipated difficult intubation (81.4% [95% CI,
74.0%– 89.0%]). Both the model described by Arné et
al. (20) and that described by Wilson et al. (1) had
lower sensitivity (54.6% [95% CI, 45.0%– 65.0%] and
Figure 1. Receiver operating characteristic (ROC) curves depicting 40.2% [95% CI, 30.0%–50.0%], respectively. However,
the relationship between the sensitivity and specificity of the three
models of predicting difficult intubation that we tested and our new
the specificity of both the Arné model (94.9% [95% CI,
model. The ROC AUCs show the discriminating power of the 90.0%–99.0%]) and Wilson model (92.8% [95% CI,
models. 88.0%–98.0%]) was significantly higher (P ⬍ 0.0001)
than that of the Naguib model (72.2% [95% CI, 63.0%–
predicting difficult intubation, the laryngoscopy and 81.0%]). Both the Naguib and Arné models were sig-
intubation would be easy if the numerical value (l) in nificantly more accurate at correctly identifying easy
the equation is less than zero (i.e., negative) but diffi- or difficult intubations (P ⫽ 0.01) than the Wilson
cult if the numerical value (l) is more than zero (i.e., model was (76.8%, 74.7%, and 66.5%, respectively).
positive). The ideal model for prediction of difficult intuba-
The posterior probability of group membership for tion would have perfect sensitivity and specificity.
each patient was used to compare the model predic- Sensitivity and specificity are dependent on each oth-
tion with the actual outcome. This new model cor- er: an increase in one of them usually results in a
rectly predicted 84% (163 of 194) of the cases. The decrease in the other. High specificity may also in-
sensitivity, specificity, positive predictive value, and crease the positive predictive value despite low sensi-
negative predictive value of this model were 82.5% tivity, as seen with the Wilson and Arné models in this
ANESTH ANALG TECHNOLOGY, COMPUTING, AND SIMULATION NAGUIB ET AL. 823
2006;102:818–24 COMPARISON OF DIFFICULT TRACHEAL INTUBATION MODELS

Table 6. Risk Factors that Correlated with the Predication prevalence of unanticipated difficult tracheal intuba-
of Difficult Intubation as Identified by Multivariate tion is small. Second, a false-positive result increases
Analysis in the New Model the potential for the serious consequence of failed
Variable ␹2 (1 DF) Odds ratio tracheal intubation.
P-value In the present study, univariate differences between
Thyromental distance 0.0218 0.630 the difficult intubation and control groups in the inter-
Mallampati score ⬍0.0001 12.827 incisor gap, thyromental distance, thyrosternal distance,
Interincisor gap 0.0005 0.318 neck circumference, and Mallampati score were noted.
Height 0.0118 1.044 The most popular clinical test for predicting the ease of
Mallampati score was dichotomized such that a Mallampati score of 1 or tracheal intubation is the Mallampati test (16). Because
2 was scored as 0 and a Mallampati score of 3 or 4 was scored as 1. DF ⫽ difficult laryngoscopy is a multifactorial problem, clearly
degrees of freedom.
no simple predictive test can be used alone. Simple bed-
side tests such as the Mallampati test (16,29), thyromen-
study. A more pressing question seems to be whether tal distance measurement (18), and sternomental dis-
sensitivity and specificity are equally important. Clin- tance measurement (30,31) have been found to be of
ical models used to predict difficult tracheal intuba- limited use in predicting difficult laryngoscopy when
tion have different trade-offs in optimizing sensitivity performed alone. Effective prediction requires a combi-
and optimizing specificity. We believe that the pur- nation of tests (4,32). A recent meta-analysis found the
pose of any such model should be detection of as combination of the Mallampati test and thyromental
many patients with a difficult airway as possible to distance to be the most accurate predictors of difficult
minimize the potentially serious consequences of un- intubation; however, this combination has a very low
anticipated difficult tracheal intubation. To that end, a sensitivity of 36% (95% CI, 14%–59%) (4).
model with high sensitivity, rather than high specific- We developed a new clinical prediction model that
ity, is required. A model with high sensitivity, low considers the thyromental distance, Mallampati score,
specificity, and low positive predictive value (as seen interincisor gap, and height. This model is 82.5% sen-
with the Naguib model) would incorrectly classify sitive and 85.6% specific with an AUC of 0.90. Height
patients as having a difficult airway. This would prob- was found to be significantly correlated with both
ably increase the financial and emotional costs for the interincisor gap (P ⬍ 0.0001) and thyromental distance
(P ⫽ 0.0007). The significance of height as a predictor
patients when, for example, an alternate intubation
of difficult intubation was addressed previously by
technique such as awake fiberoptic-guided intubation,
Schmitt et al. (33). They reported that the ratio of
is used. However, these costs may only be a fraction of
height to thyromental distance was a more sensitive
those that accompany the potentially serious outcome
indictor of difficult intubation than the thyromental
of unanticipated difficult tracheal intubation. There-
distance alone (33). We considered a model that in-
fore, the sensitivity of a prediction model is more
cluded the ratio of height to thyromental distance,
important than the specificity and should be weighted
which yielded identical results to the new model. For
more heavily when determining which model to use. the sake of parsimony, we chose to include only height
The value of the simplified risk index used in this in our model instead of the ratio of height to thyro-
study (⬎11) was the value recommended by Arné et al. mental distance, as suggested by Schmitt et al. (33).
(20). In their original description of their model, Naguib The new model must be prospectively validated.
et al. (21) reported a sensitivity and specificity of 95% The incidence of unanticipated difficult tracheal intu-
and 91%, respectively, whereas Arné et al. (20) reported bation in this study (0.13%) is less frequent than the
a sensitivity and specificity of 94% and 93%, respectively. range of 1%–18% reported by others (1–3). Also, the
Of note is that in the present study both models had a incidence of impossible tracheal intubation in our study
lower sensitivity than previously demonstrated. (0.05%) is at the low end of the range of 0.05%– 0.35%
Oates et al. (17) evaluated the Wilson risk sum reported previously (5,6). The number of patients en-
(score ⱖ2) in 675 cases. They reported a positive pre- rolled in the first 15 months of our study was 44% of the
dictive value of 8.9% with a low sensitivity (42%) and total number of patients (86 of 194). However, the num-
high specificity (92%). Using the same threshold, ber of patients enrolled decreased dramatically over ap-
Yamamoto et al. (28) reported that the Wilson risk sum proximately the next four years, suggesting a possible
yielded a low positive predictive value (5.9%), low Hawthorne effect. The authors feel that this study may
sensitivity (55.4%), and high specificity (86.1%). Simi- have increased practitioner awareness of difficult air-
larly, Siddiqi and Kazi (22) reported that both Wilson ways in patients presenting for surgery and prompted
risk sum (score ⱖ2) and Mallampati classification more aggressive use of alternate airway-management
have a similar sensitivity of 42% but different positive techniques, leading to a decrease in the incidence of
predictive values of 11% and 5%, respectively. A unanticipated difficult intubation. A Hawthorne effect
higher threshold is preferred for two reasons. First, the (identified observer effect) is defined as the tendency of
824 TECHNOLOGY, COMPUTING, AND SIMULATION NAGUIB ET AL. ANESTH ANALG
COMPARISON OF DIFFICULT TRACHEAL INTUBATION MODELS 2006;102:818–24

individuals to improve their behaviors or performance 12. Johnson KG, Hood DD. Esophageal perforation associated with
endotracheal intubation. Anesthesiology 1986;64:281–3.
when they know that they are under observation (34,35). 13. Biboulet P, Aubas P, Dubourdieu J, et al. Fatal and non fatal
The authors also realize they were dependent upon cardiac arrests related to anesthesia. Can J Anaesth 2001;48:
many individual practitioners’ assessment and self- 326–32.
reporting of difficult airways. This is a known limitation 14. Rose DK, Cohen MM. The airway: problems and predictions in
18,500 patients. Can J Anaesth 1994;41:372–83.
of voluntary reporting techniques for critical incidents in 15. Practice guidelines for management of the difficult airway: a
health care (36). report by the American Society of Anesthesiologists Task Force
A potential limitation of a matched case-controlled on Management of the Difficult Airway. Anesthesiology 1993;
study (such as this study) is the possibility that some 78:597–602.
16. Mallampati SR, Gatt SP, Gugino LD, et al. A clinical sign to
segments of population may not be adequately repre- predict difficult tracheal intubation: a prospective study. Can
sented in the study participants. Another limitation of Anaesth Soc J 1985;32:429–34.
our study was that it was not a truly prospective 17. Oates JD, Macleod AD, Oates PD, et al. Comparison of two
methods for predicting difficult intubation. Br J Anaesth 1991;
study. It can be best described as a quasi-prospective 66:305–9.
evaluation of three models for the prediction of unan- 18. Butler PJ, Dhara SS. Prediction of difficult laryngoscopy: an
ticipated difficult intubation, because patients were assessment of the thyromental distance and Mallampati predic-
identified, recruited, and examined after attempted tive tests. Anaesth Intensive Care 1992;20:139–42.
19. el-Ganzouri AR, McCarthy RJ, Tuman KJ, et al. Preoperative
intubation. However, we do not believe that this airway assessment: predictive value of a multivariate risk index.
would have a significant impact on our results. Anesth Analg 1996;82:1197–204.
In conclusion, our study is the first to provide an 20. Arné J, Descoins P, Fusciardi J, et al. Preoperative assessment for
difficult intubation in general and ENT surgery: predictive
evidence-based foundation for selection of the most value of a clinical multivariate risk index. Br J Anaesth 1998;80:
sensitive model for prediction of unanticipated diffi- 140–6.
cult tracheal intubation. We confirmed the high sensi- 21. Naguib M, Malabarey T, AlSatli RA, et al. Predictive models for
tivity of the Naguib model but failed to do so for the difficult laryngoscopy and intubation: a clinical, radiologic and
three-dimensional computer imaging study. Can J Anaesth
Arné and Wilson models. We also created a new 1999;46:748–59.
model for predicting unanticipated difficult intuba- 22. Siddiqi R, Kazi WA. Predicting difficult intubation: a compari-
tion, although it has not yet been prospectively tested. son between Mallampati classification and Wilson risk-sum. J
This model may be more sensitive and specific than Coll Physicians Surg Pak 2005;15:253–6.
23. Edge WG, Whitwam JG. Chondro-calcinosis and difficult intu-
any of the models currently used to predict difficult bation in acromegaly. Anaesthesia 1981;36:677–80.
intubation. 24. Calder I, Calder J, Crockard HA. Difficult direct laryngoscopy in
patients with cervical spine disease. Anaesthesia 1995;50:756–63.
25. Hanley JA, McNeil BJ. The meaning and use of the area under
a receiver operating characteristic (ROC) curve. Radiology 1982;
References 143:29–36.
1. Wilson ME, Spiegelhalter D, Robertson JA, Lesser P. Predicting 26. Somers RH. A new asymmetric measure of association for or-
difficult intubation. Br J Anaesth 1988;61:211–6. dinal variables. Am Sociol Rev 1962;27:799–811.
2. Benumof JL. Management of the difficult adult airway: with 27. DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the
special emphasis on awake tracheal intubation. Anesthesiology areas under two or more correlated receiver operating charac-
1991;75:1087–110. teristic curves: a nonparametric approach. Biometrics 1988;44:
3. Tse JC, Rimm EB, Hussain A. Predicting difficult endotracheal 837–45.
intubation in surgical patients scheduled for general anesthesia: 28. Yamamoto K, Tsubokawa T, Shibata K, et al. Predicting difficult
a prospective blind study. Anesth Analg 1995;81:254–8. intubation with indirect laryngoscopy. Anesthesiology 1997;86:
4. Shiga T, Wajima Z, Inoue T, Sakamoto A. Predicting difficult 316–21.
intubation in apparently normal patients: a meta-analysis of 29. Cattano D, Panicucci E, Paolicchi A, et al. Risk factors assess-
bedside screening test performance. Anesthesiology 2005;103: ment of the difficult airway: an Italian survey of 1956 patients.
429–37. Anesth Analg 2004;99:1774–9.
5. Cormack RS, Lehane J. Difficult tracheal intubation in obstetrics. 30. Savva D. Prediction of difficult tracheal intubation. Br J Anaesth
Anaesthesia 1984;39:1105–11. 1994;73:149–53.
6. Samsoon GL, Young JR. Difficult tracheal intubation: a retro- 31. Al Ramadhani S, Mohamed LA, Rocke DA, Gouws E. Sterno-
spective study. Anaesthesia 1987;42:487–90. mental distance as the sole predictor of difficult laryngoscopy in
7. Benumof JL. Difficult laryngoscopy: obtaining the best view. obstetric anaesthesia. Br J Anaesth 1996;77:312–6.
Can J Anaesth 1994;41:361–5. 32. Bainton CR. Difficult intubation: what’s the best test? Can J
8. Confidential enquiry into maternal and child health: why moth- Anaesth 1996;43:541–3.
ers die 2000 –2002—the sixth report of the confidential enquiries 33. Schmitt HJ, Kirmse M, Radespiel-Troger M. Ratio of patient’s
into maternal death in the United Kingdom. London: RCOG height to thyromental distance improves prediction of difficult
Press, 2004. laryngoscopy. Anaesth Intensive Care 2002;30:763–5.
9. King TA, Adams AP. Failed tracheal intubation. Br J Anaesth 34. Parsons HM. What happened at Hawthorne? Science 1974;183:
1990;65:400–14. 922–32.
10. Caplan RA, Posner KL, Ward RJ, Cheney FW. Adverse respira- 35. De Amici D, Klersy C, Ramajoli F, et al. Impact of the Haw-
tory events in anesthesia: a closed claims analysis. Anesthesiol- thorne effect in a longitudinal clinical study: the case of anes-
ogy 1990;72:828–33. thesia. Control Clin Trials 2000;21:103–14.
11. Boelle PY, Garnerin P, Sicard JF, et al. Voluntary reporting 36. Figueiras A, Tato F, Fontainas J, et al. Physicians’ attitudes
system in anaesthesia: is there a link between undesirable and towards voluntary reporting of adverse drug events. J Eval Clin
critical events? Qual Health Care 2000;9:203–9. Pract 2001;7:347–54.